d_model

d_model builds reinforcement learning environments designed for open-ended interpretability tasks, training agents to conduct genuine research rather than game reward systems. The lab partners with frontier AI labs to help transform models into capable interpretability and alignment researchers and publishes research on model steering and code understanding.

The company views interpretability as central to making models safer and more steerable and treats alignment as an unsolved high-stakes problem. It was founded by Anish Tondwalkar and Daniel Moon and is based in San Francisco.